Pentesting

OAuth to Account takeover

This information was taken from https://medium.com/a-bugz-life/the-wondeful-world-of-oauth-bug-bounty-edition-af3073b354c1

“Quick” Primer

There are a couple different versions, as well as grant types to consider when we talk about OAuth. To read about these, I recommend reading through https://oauth.net/2/ to get a baseline understanding. In this article, we will be focusing on the most common flow that you will come across today, which is the OAuth 2.0 authorization code grant type. In essence, OAuth provides developers an authorization mechanism to allow an application to access data or perform certain actions against your account, from another application (the authorization server).

For example, let’s say website https://yourtweetreader.com has functionality to display all tweets you’ve ever sent, including private tweets. In order to do this, OAuth 2.0 is introduced. https://yourtweetreader.com will ask you to authorize their Twitter application to access all your Tweets. A consent page will pop up on https://twitter.com displaying what permissions are being requested, and who the developer requesting it is. Once you authorize the request, https://yourtweetreader.com will be able to access to your Tweets on behalf of you. Now, this was very high level, and there’s some complexity here. Taking this example, here’s a bit more details on the particular elements which are important to understand in an OAuth 2.0 context:

resource owner: The resource owner is the user/entity granting access to their protected resource, such as their Twitter account Tweets

resource server: The resource server is the server handling authenticated requests after the application has obtained an access token on behalf of the resource owner . In the above example, this would be https://twitter.com

client application: The client application is the application requesting authorization from the resource owner. In this example, this would be https://yourtweetreader.com.

authorization server: The authorization server is the server issuing access tokens to the client application after successfully authenticating the resource owner and obtaining authorization. In the above example, this would be https://twitter.com

client_id: The client_id is the identifier for the application. This is a public, non-secret unique identifier.

client_secret: The client_secret is a secret known only to the application and the authorization server. This is used to generate access_tokens

response_type: The response_type is a value to detail which type of token is being requested, such as code

scope: The scope is the requested level of access the client application is requesting from the resource owner

redirect_uri: The redirect_uri is the URL the user is redirected to after the authorization is complete. This usually must match the redirect URL that you have previously registered with the service

state: The state parameter can persist data between the user being directed to the authorization server and back again. It’s important that this is a unique value as it serves as a CSRF protection mechanism if it contains a unique or random value per request

grant_type: The grant_type parameter explains what the grant type is, and which token is going to be returned

code: This code is the authorization code received from the authorization server which will be in the query string parameter “code” in this request. This code is used in conjunction with the client_id and client_secret by the client application to fetch an access_token

access_token: The access_token is the token that the client application uses to make API requests on behalf of a resource owner

refresh_token: The refresh_token allows an application to obtain a new access_token without prompting the user

Well, this was meant to be a quick primer but it seems with OAuth, you can’t simply give a brief description. Putting this all together, here is what a real OAuth flow looks like:

  1. You visit https://yourtweetreader.com and click the “Integrate with Twitter” button.

  2. https://yourtweetreader.com sends a request to https://twitter.com asking you, the resource owner, to authorize https://yourtweetreader.com’s Twitter application to access your Tweets. The request will look like:

https://twitter.com/auth
?response_type=code
&client_id=yourtweetreader_clientId
&redirect_uri=https%3A%2F%2Fyourtweetreader.com%2Fcallback
&scope=readTweets
&state=kasodk9d1jd992k9klaskdh123

3. You will be prompted with a consent page:

4. Once accepted, Twitter will send a request back to the redirect_uri with the code and state parameters:

https://yourtweetreader.com?code=asd91j3jd91j92j1j9d1&state=kasodk9d1jd992k9klaskdh123

5. https://yourtweetreader.com will then take that code , and using their application’s client_id and client_secret , will make a request from the server to retrieve an access_token on behalf of you, which will allow them to access the permissions you consented to:

POST /oauth/access_token
Host: twitter.com
...{"client_id": "yourtweetreader_clientId", "client_secret": "yourtweetreader_clientSecret", "code": "asd91j3jd91j92j1j9d1", "grant_type": "authorization_code"}

6. Finally, the flow is complete and https://yourtweetreader.com will make an API call to Twitter with your access_token to access your Tweets.

Bug Bounty Findings

Now, the interesting part! There are many things that can go wrong in an OAuth implementation, here are the different categories of bugs I frequently see:

Weak redirect_uri configuration

This is probably one of the more common things everyone is aware of when looking for OAuth implementation bugs. The redirect_uri is very important because sensitive data, such as the code is appended to this URL after authorization. If the redirect_uri can be redirected to an attacker controlled server, this means the attacker can potentially takeover a victim’s account by using the code themselves, and gaining access to the victim’s data.

The way this is going to be exploited is going to vary by authorization server. Some will only accept the exact same redirect_uri path as specified in the client application, but some will accept anything in the same domain or subdirectory of the redirect_uri .

Depending on the logic handled by the server, there are a number of techniques to bypass a redirect_uri . In a situation where a redirect_uri is https://yourtweetreader.com/callback, these include:

  • Open redirects: https://yourtweetreader.com/callback?redirectUrl=https://evil.com

  • Path traversal: https://yourtweetreader.com/callback/../redirect?url=https://evil.com

  • Weak redirect_uri regexes: https://yourtweetreader.com.evil.com

  • HTML Injection and stealing tokens via referer header: https://yourtweetreader.com/callback/home/attackerimg.jpg

Improper handling of state parameter

This is by far the most common issue I see in OAuth implementations. Very often, the state parameter is completely omitted or used in the wrong way. If a state parameter is nonexistent, or a static value that never changes, the OAuth flow will very likely be vulnerable to CSRF. Sometimes, even if there is a state parameter, the application might not do any validation of the parameter and an attack will work. The way to exploit this would be to go through the authorization process on your own account, and pause right after authorizing. You will then come across a request such as:

https://yourtweetreader.com?code=asd91j3jd91j92j1j9d1

After you receive this request, you can then drop the request because these codes are typically one-time use. You can then send this URL to a logged-in user, and it will add your account to their account. At first, this might not sound very sensitive since you are simply adding your account to a victim’s account. However, many OAuth implementations are for sign-in purposes, so if you can add your Google account which is used for logging in, you could potentially perform an Account Takeover with a single click as logging in with your Google account would give you access to the victim’s account.

You can find an example about this in this CTF writeup and in the HTB box called Oouch.

I’ve also seen the state parameter used as an additional redirect value several times. The application will use redirect_uri for the initial redirect, but then the state parameter as a second redirect which could contain the code within the query parameters, or referer header.

One important thing to note is this doesn’t just apply to logging in and account takeover type situations. I’ve seen misconfigurations in:

  • Slack integrations allowing an attacker to add their Slack account as the recipient of all notifications/messages

  • Stripe integrations allowing an attacker to overwrite payment info and accept payments from the victim’s customers

  • PayPal integrations allowing an attacker to add their PayPal account to the victim’s account, which would deposit money to the attacker’s PayPal

Assignment of accounts based on email address

One of the other more common issues I see is when applications allow “Sign in with X” but also username/password. There are 2 different ways to attack this:

  1. If the application does not require email verification on account creation, try creating an account with a victim’s email address and attacker password before the victim has registered. If the victim then tries to register or sign in with a third party, such as Google, it’s possible the application will do a lookup, see that email is already registered, then link their Google account to the attacker created account. This is a “pre account takeover” where an attacker will have access to the victim’s account if they created it prior to the victim registering.

  2. If an OAuth app does not require email verification, try signing up with that OAuth app with a victim’s email address. The same issue as above could exist, but you’d be attacking it from the other direction and getting access to the victim’s account for an account takeover.

Disclosure of Secrets

It’s very important to recognize which of the many OAuth parameters are secret, and to protect those. For example, leaking the client_id is perfectly fine and necessary, but leaking the client_secret is dangerous. If this is leaked, the attacker can potentially use the trust and identity of the trusted client application to steal user access_tokens and private information/access for their integrated accounts. Going back to our earlier example, one issue I’ve seen is performing this step from the client, instead of the server:

5. https://yourtweetreader.com will then take that code , and using their application’s client_id and client_secret , will make a request from the server to retrieve an access_token on behalf of you, which will allow them to access the permissions you consented to.

If this is done from the client, the client_secret will be leaked and users will be able to generate access_tokens on behalf of the application. With some social engineering, they can also add more scopes to the OAuth authorization and it will all appear legitimate as the request will come from the trusted client application.

Closing

There’s plenty of other attacks and things that can go wrong in an OAuth implementation, but these are some of the more common ones that you will see. These misconfigurations are surprisingly common, and a very large quantity of bugs come from these. I intended to keep the “Quick Primer” rather short, but quickly realized all of the knowledge was necessary for the rest of the post. Given this, it makes sense that most developers aren’t going to know all the details for implementing securely. More often than not, these issues are high severity as it involves private data leak/manipulation and account takeovers. I’d like to go into more detail in each of these categories at some point, but wanted this to serve as a general introduction and give ideas for things to look out for!

OAuth providers Race Conditions

If the platform you are testing is an OAuth provider read this to test for possible Race Conditions.